0.20.8

What's Changed

• Fix dialect for JAX mhlo -> stablehlo. by @ameya98 in #85
• Add support for signals on SO3. by @ameya98 in #84
• Fix SO3 sampling. by @ameya98 in #86
• Don't pre-multiply by angle measure. by @ameya98 in #87
• Add tests. by @ameya98 in #88
• Minor fixes for SO3Signal by @ameya98 in #89
• Fix multiplication. by @ameya98 in #90
• Add argmax for signals on SO3 by @ameya98 in #91
• Fix formatting. by @ameya98 in #92
• Add apply() for SO3Signal. by @ameya98 in #94
• Add division and negation for SO3Signal. by @ameya98 in #95
• Add static fields to e3nn_jax.equinox.Linear by @teddykoker in #96
• Add equinox modules to docs by @teddykoker in #97

New Contributors

• @teddykoker made their first contribution in #96

Full Changelog: 0.20.7...0.20.8

0.20.7

What's Changed

• Cache cg coefficients by @mitkotak in #73
• Updated citation by @mitkotak in #76
• Add custom initializers and instructions in Linear by @ameya98 in #78
• Fix documentation build errors by @ameya98 in #79

New Contributors

• @mitkotak made their first contribution in #73

Full Changelog: 0.20.6...0.20.7

2024-01-26

Added

• e3nn.where function
• Add optional mask argument in e3nn.flax.BatchNorm

Changed

• replace jnp.ndarray by jax.Array

2024-01-05

Added

• e3nn.ones and e3nn.ones_like functions
• e3nn.equinox submodule

Fixed

• python 3.9 compatibility

Thanks to @ameya98, @SauravMaheshkar and @pabloferz

2023-12-24

Fixed

• Fix pyproject.toml, the documenation build was broken. Thanks to @SauravMaheshkar!

Added

• Support for s2fft in e3nn.to_s2grid and e3nn.from_s2grid, thanks to @ameya98!
• Add a special case implementation for e3nn.scatter_mean when map_back and nel is not None.

2023-11-17

Added

• e3nn.flax.BatchNorm
• e3nn.scatter_mean
• Add e3nn.utils.vmap also directly to e3nn module: e3nn.vmap

2023-09-25

Added

• with_bias argument to e3nn.haiku.MultiLayerPerceptron and e3nn.flax.MultiLayerPerceptron

Fixed

• Improve compilation speed and stability of s2grid for large lmax (use is_normalized=True in lpmn_values)

2023-09-13

Changelog

Changed

• Add back the optimizations with the lazy ._chunks that was removed in 0.19.0

2023-09-09

Highlight

tl;dr Mostly fix the issue #38

In version 0.19.0, I removed the lazy _list attribute of IrrepsArray to fix the issues from tree_util, grad and vmap.
In this version (0.20.0) I found a way to put back that lazy attribute, now called _chunks, in a way that does not interfere with tree_util, grad and vmap. _chunks is tropped when using tree_util, grad and vmap unless you use e3nn.vmap.

ChangeLog

Added

• e3nn.Irreps.mul_gcd
• e3nn.IrrepsArray.extend_with_zeros to extend an array with zeros, can be useful for residual connections

Changed

• rewrite e3nn.tensor_square to be simpler (and faster?)
• use jax.scipy.special.lpmn_values to implement e3nn.legendre. Faster on GPU and supports reverse-mode differentiation.
• [BREAKING] Change the output format of e3nn.legendre!

Fixed

• Add back a lazy ._chunks in e3nn.IrrepsArray to fix issue #38

2023-06-24

Changelog

Fixed

• Fix missing support for zero flags in e3nn.elementwise_tensor_product